Decomposition Revised of Multiport Multibond Elements in a Graph Notation hy PETER C.BREEDVELD Twente University of Technology, 217, 7500 AE Enschede, Department of Electrical Engineering, P.O. Box Netherlands ABSTRACT : Decomposition rules are derived for multiport-transformers, -resistors, -storage elements and -gyrators into l- and 2-port elements, junctions and bonds. It appears that it is useful to extend the vectorbond, or rather multibond, notation recently proposed by the author with a “multibond array”. Canonical forms are introduced on the basis of minimal realization, because decompositions of multiport elements are not unique. A new type of coupling factor (“directed coupling factor”) is introduced,for multiport-resistors and capacitors. I. Introduction Decomposition of multiport elements has a twofold purpose. It may enhance one’s insight into the characteristics of the represented system, which may be helpful in design, especially the design of transducers ofwhich the ideal form corresponds to special cases of multiport storage elements (e.g. an electrical transformer) or multiport resistors (e.g. a thermocouple). Secondly, it enables the benefits of simple simulation programs without a multiport subroutine facility [like THTSIM (l)]. Canonical decomposition even results in the most simple way of computation, which may be useful if computation time is the bottle-neck in digital simulation, These aspects will not be elaborated upon. The decomposition rules will be presented without discussing the applications of (canonical) decomposition in order to concentrate on the resemblances and differences between the algorithms. Multiport elements can be represented in several ways of which the circuit diagram representation is most often used [Fig. l(a)]. The less familiar bond graph notation (2,3) to be used in this paper has many advantages. Figure l(b) already shows some practical ones, irrespective of its meaning, i.e. it takes less effort to draw and it requires less space, which is important in the case of large and complicated systems. The representation of the concept of computational causality, i.e. the computational “direction” of the effort-variable(s) indicated by a so-called causal stroke (Fig. 3), is another important feature of this notation. It enables a simple check on unstable and algebraic loops, besides many features which are not discussed here [see, for instance, (2)] because the notation will only be used for algorithmic purposes. In (4) a proposition has been presented for an unambiguous notation of multiport systems with the use of vectorbonds. The terminology “vectorbond”, introduced by Bonderson (S), has the disadvantage that it is often associated with some directed CC,The FranklinInstitute 0016 0032/84$3.00+0.00 253 Peter C. Breedveld (a) chorocteristics lb) FIG. 1. (a) Multiport element in circuit diagram representation. (b) Multiport single-bond and multibond graph notation. element in property, while the word “vector” is used in this case in the sense of a set of bonds which can be written as a “column” or a “row”. In order to prevent this confusion from now on the terminology “multibonds”, rather than “vectorbonds”, will be used. The basic idea behind decomposition is the generation of some “analytical junction structure” (AJS) to which l- and 2-port elements are connected in such a way that, with respect to the (external) ports, this new decomposed multiport system is identical to the original multiport element (Fig. 2). In contrast to the “physical junction structure” (PJS) which represents physical connections in a system (topology), the AJS has no direct physical meaning, but its analogy to a PJS can be very illuminating. Decomposition rules will be derived successively for the multiport transformer (notation: “TF” in “shaded characters” if drawn [cf. Fig. 4(a)], or “MP TF” in a printed text), the multiport resistor (MP R) or, in the non-isothermal case, the irreversible transducer (MP RS) (6), the multiport capacitor (MP C) or multiport inertance (MP I) and the multiport gyrator (MP GY). All MPs can be modulated, i.e. can have a variable constitutive parameter, except for the MP C and MP I, because modulation would violate the energy conservation of these elements. Modulation of a multiport element is indicated by an additional M in the mnemonic code: MP FIG. 2. Decomposition 254 of a multiport element into an analytical junction port elements. structure and l- or 2- Journal of the Franklin Institute Pergamon Press Ltd. Decomposition corresponding constitutive of Multiport Elements relation: e,=Xf, _ +-+qp:x -i FIG. 3. Causally augmented multiport element. MTF, MP MR, etc. Multiport effort- and flow-sources only occur in decomposed form, i.e. as an array of l-port sources (4), because their constitutive relation is not a function of the independent powervariable of the port(s). The decomposition of the MP TF is treated first, because this decomposition will be used in almost all other decompositions (Section II). Next the MP R is discussed (Section III), followed by the decomposition of the MP C (or MP I) which, in the linear case, is similar to the decomposition of the MP R [Section IV(a)], but more complicated in the nonlinear case [Section IV(b)], where “internal modulation” is required. In Section III a coupling factor is introduced for MP Rs, which is also used in Section IV(a) for linear MP Cs. These so-called “directed coupling factors”, in which the sign information of the cross-effects is preserved, are used to formulate the canonical forms of 2- and 3-ports. Finally the different characteristics of the possible alternatives for the decomposition of the MP GY justify its treatment last (Section V). In all cases it has to be assumed that the constitutive relations can be written in matrix form, corresponding to the causality assignment of the MP (cf. Fig. 3). The matrix does not necessarily have constant coefficients in all cases, but the coefficients have to be real, because they represent parameters of physical devices (only the timedomain is considered). Except for the MP TF, for which the constitutive equation directly suggests its (immediate canonical) decomposition, two decomposition techniques can be applied to the multiport elements resulting in two different canonical forms. The first technique is based on the congruence properties of matrices (“congruence canonical form”). The second canonical form will be called “immediate canonical form”, because some decomposition is postulated (usually on the basis of the constitutive equation or the congruence canonical form) and checked to satisfy the same constitutive equations as the original multiport element. II. Decomposition By definition relation : of Multiport a (modulated) Transformers multiport transformer has the following [;:I=[_“, Y][Z:] constitutive (1) where e and fare vectors of conjugate power variables effort and flow respectively (2), T is an m x n-dimensional matrix with real coefficients, which are not necessarily constant : T = T(.) and TT is the transpose of T. Its multibond graph representation Vol. 318, No. 4, pp. 253-273, Printed III Great Britam October 1984 255 Peter C. Breedveld (a) (b) Cc) (d) n % 2 :m & 0 I-; 1 n FIG. 4.(a)(Modulated) multiport transformer. (b) Causally augmented (modulated) multiport transformer. (c) Single-bond decomposition of a multiport transformer. (d) Multibond decomposition of a multiport transformer. is shown in Fig. 4(a). If Eq. (1) is written in the form e, = TT(.)e, -f* = T(.)f, the more familiar representation shown in Fig. 4(b) is obtained. Decomposition of this MP TF corresponds to the decomposition of the products of the matrices TT and T with the vectors e, and f, respectively, into scalar products (to be represented by 2-port transformers) and summations (to be represented by l- and O-junctions). This (immediate) decomposition is canonical, because the number of elements (2P TFs) is equal to the number of independent constitutive parameters (n x m). 256 Journal of the Franklin Institute Pergamon Press Ltd. Decomposition According of Multiport Elements to -f’2 = c ljif'; i=l which can be written ei,j = -fir’ T,,ei.i ,’ 2 = Tjif’;’ (TF) with ey = ,j Vi 2 (O-junction) f’, = i$l/Y and n ei = C ei,j j=l (l-junction). fl;j=fi vj Figure 4(c) represents this decomposition in a conventional, “single-bond” way, while in Fig. 4(d) the multibond notation and terminology introduced in (4) is used in order to “compress” the notation in connection with decompositions of other multiport elements. Although array junctions, a 2-port transformer array and the direct sum are used, Fig. 4(d) still contains dotted lines representing arrays of multibonds, which in turn are ofcourse multibonds. In addition to (4) the notation of Fig. 5(a) is proposed for this situation and called “multibond array”. In anticipation of situations occurring in decompositions of other MPs, the parameter m, representing the number of multibonds, is allowed to be a function of the junction index i (1 < i < n). If the multibond array notation is applied to the MP TF, Fig. 4(d) becomes Fig. 5(b). For the crossing matrix of the first direct sum the unit matrix can be taken (default value of the constitutive matrix of the direct sum) (4), while the crossing matrix K of the second direct sum has the form [Fig. 5(b)] : for {k, 1>E [l,(m x n)] K,,=l ifk=((r-l)n+s) and /=((s-l)n+r) V{r,s)~[l,m] and K,, = 0 otherwise. It is important to note that the decomposition techniques presented are independent of the multibond notation. However, if one is accustomed to this notation its Vol. 318, No. 4, pp, 253-273, Printed m Great Brttam October 1984 257 Peter C. Breedveld (0) n _j usually where in the m form: ,! is a 0- or l-junction array and i iz the junction index (b) FIG. 5. (a) Definition of a multibond array. (b) Multibond decomposition transformer with the use of multibond arrays. of a multiport and transparency facilitate the representation of the basic features of the decomposition process. The fact will be used in the decompositions of the other multiport elements that some MP X characterized by a matrix X (e.g. e = Xf) and “seen through” a MP TF with constitutive matrix T[Fig. 6(a)] can be “transformed over the MP TF”. This yields a new MP Y with constitutive matrix Y = TT XT [Fig. 6(b)]. In other words, Fig. 6(a) and (b) are equivalent from the viewpoint of the (external) multiport 1. This compactness (a) (b) FIG. 6. (a) Multiport element “seen through” a multiport transformer. (b) Resulting multiport element of a “transformation of a multiport element over a multiport transformer” (“composition”). 258 Journal of the Franklin Institute Pergamon Press Ltd. Decomposition can easily be verified by writing MPX: e, = Xf, MP TF: e, = TTe, fi = Tf, MPY: of Multiport Elements : out the equations ] + e, = TTXTf, + I e, = Yf, Y = TTXT. I Finally, it is remarked that a MP TF is sometimes called weightedjunction structure (WJS), a structure of bonds and junctions (junction structure, JS) also containing transformers (“weighting factors”) (7), especially if its similarity to a “physical” transformer is not obvious. of&f&tip0 Hi. lhxomph In the constitutive relation fl Resistors of a multiport resistor [Fig. 7(a)] e = Rf or f = R-‘e the constitutive matrix R is symmetric according to Onsager reciprocity (6,8) R = RT and positive-definite according to the second law of thermodynamics definite if the degenerate (dissipationless) case is included] i.e. fTRf>O [positive semi- Vf#O. A linearized MP R may have operating points where R is not positive-definite without violating the second law of thermodynamics. These cases have to be excluded and studied individually. A positive-definite, real, symmetric matrix R is congruent (even orthogonally similar, UT = U-‘) to a diagonal matrix of which the elements are the eigenvalues (characteristic roots) of R : R=U=AU where A = diag {A,, . . . ,&,. . .}. This suggests a “spectral decomposition” as presented in Fig. 7(b), where the MP TF can be decomposed according to Section II. Although this matrix transformation is usually called a canonical transformation, the corresponding decomposed form will not be considered canonical because the number of elements (n2 TFs and n Rs) is larger than the number of independent parameters of R ((n/2)(n+ 1)) and thus not minimal. Therefore another decomposition will be studied, based on the congruence of a positive-definite, real, symmetric matrix R with a unit matrix Z : R = TTZT= Vol. 318, No. 4, pp. 253-273, Printed in Great Britain October 1984 TT?: 259 Peter C. Breedveld (cl FIG. 7. (a) Multiport resistor. (b) Spectral decomposition of a multiport resistor. (c) Singlebond congruence canonical form. (d) Multibond congruence canonical form. The matrix T and accordingly the corresponding decomposition are not unique however. Two special cases exist, of which the first, where T is symmetric : TT= T R = T2 with has the same form considered canonical. as the first decomposition and consequently The second, where T is triangular : Tj=O cannot be fori>j approximates the minimal number of elements more closely ((42) (n + 1) TFs, n Rs), but the n l-port Rs are unity and do not have a parameter related to the matrix R. Because of the triangular structure of T, the n TFs corresponding to the diagonal coefficients of T, can be eliminated by transforming the n unit Rs (for the 260 Journalorthe Franklin Institute Pergamon Press Ltd. Decomposition transformation II) : of elements over transformers, see concluding of Multiport remarks Elements of Section The corresponding minimal decomposition which accordingly is considered to be the congruence canonical form is presented in Fig. 7(c) and (d) in single-bond and multibond notation respectively. By solving the resulting equations in tij and di in terms of the coefficients of the matrix R, the decomposition is completed. A bond graph algorithm to obtain the congruence canonical decomposition of an n-port resistor with effort causality using the conventional notation is presented in Fig. S(a). The decomposed 2- and 3-port resistors in Fig. 8(b), for instance, are obtained with the use of this algorithm. The constitutive parameters are expressed in terms of the coefficients of the matrix R and a new type of coupling factor “k “k = Rij/Rii # “k = Rji/Rjj = Rij/Rjj with 0 Q “kj’k < 1 because R is positive-definite. This will be called the directed coupling factor in contrast coupling factor for 2-ports k : to the conventional The advantage of the directed coupling factor is that the information on the sign of Rij is not lost. In case of flow-causality (f = R-‘e) the same algorithm can be used except for a dualization of the junctions [Fig. 8(c)]. The bond graph representation of the congruence canonical form [Fig. 7(c)] suggests an immediate canonical decomposition which cannot be found from the equations. Suppose that each port is modeled by a l-junction with a (1P) R connected to it, whereas each l-junction is connected to the other l-junctions via a Ojunction with a (1P) R connected to it. Then Fig. 8(d) shows the decomposition of a 2and a 3-port R.7 The resulting equations are : 2P: e, = (Rl+R12Vl+R12f2 e2 = R12fl +(JG +Uh t The 2-port in Fig. 8(d) corresponds to a resistive T-network in circuit theory. In case of a 2-port R it is even possible to enumerate all possible canonical decompositions (i.e. with three elements). It can easily be checked that the immediate canonical form (T-network) and its dual (n-network) and the congruent canonical form (its dual corresponds to interchanging the ports) constitute all possible canonical decompositions of a 2-port R. Vol. 318, No. 4, pp. 253-273, Printed in Great Britain October 1984 261 Peter C. Breedveld with RI = R,,-R,, R, = Rx-R,2 3P: (Rl+R12+R13)fl+R12f2+R13f3 el = e2 =R12fl+(R2+R12+R23)_f2+R23.f3 R13.fl+R23fZ+(R3+R13+R23)f3 e3 = (a) -Draw n times -Simplify for -Connect right -The from i-1 left and to i-n each l-junction it by a 2-port of transformation right: into: to ratio of determined by means (e.g. the transformer 1”‘): l-junction transformation to of the ratio each O-junction to the transformer: jth the the transformers labels which of the connects O-junction (i<j) can be junctions the has ith a Til FIG. 8. (a) Bond graph algorithm to determine the congruence canonical form of a multiport resistor. (b) 2- and 3-port congruence canonical forms. (c) 2-port congruence canonical form with “conductive causality”. (d) 2- and 3-port immediate canonical forms. 262 Joumal of the Franklin institute Pergamon Press Ltd. Decomposition of Multiport Elements e (b) t f: 7 I -TF+ o \ % f, t J Rd d, 1 R,d, = R,, d, (l- 12k2’k) = R22 t -“k e 8 I -TF- I f : R.d, Rd, Rd, d, - R,,(I- "k *'k-13ka$_ *% '\+2 '2k23k3'~)~(1_23k32k)-' d2 = Rz2(l-23k “k) d, t12 t13 t23 = R33 = ( *‘k- =k3’k) (I_ 23k32k)-’ = “k s =k (d) _iiyi+ (cl e, f, RR, RR,, RR, RR,, RR, f* e, ~oc---ITF-I- 1 1 R ’ I T R RR, 0~1'----70 J es!3J R.4, RR,, FIG. 8 (continued). Vol. 318, No. 4, pp. 253-273, Printed in Great Britain October 1984 263 Peter C. Breedveld FIG. 9. Decomposition of an irreversible transducer (non-isothermal resistor). with R, = RI,-J&-R,, which can be generalized R, = R,, -RI, -IL R, = R,, --RI, - R,, for an n-port ei = (Ri+~iRij)/;+~iRiLI, Ri = R,,- 1 R,. jti These equations indeed have the general form of the constitutive equations of a MP R. If the directed coupling factor is used the constitutive parameters of the Rs connected are : to the l-junctions Ri = Rii and the parameters of the Rs connected to the O-junctions are : R..I, = R..1, = R..II “k = R ..j’k. JJ The presented decomposition rules are similar to those for modulated resistors (MP MR) and irreversible transducers (MP RS). In the latter case all l-port resistors in the decomposition are replaced by 2-port irreversible transducers RS of which the entropy producing ports are connected to a O-junction (Fig. 9). In case the resistor is modulated, the l- and 2-port elements characterized by (a function of) variable matrix coefficients become modulated l- and 2-port elements. If the MP R is nonlinear the decomposition rules also apply, because no other constraints than that the matrix R should be positive-definite, real and symmetric have been imposed (Fig. 10). The non-linearity results in variable matrix coefficients which can be represented by internal modulation, i.e. modulation by one of the conjugate powervariables of the multiport element itself [cf. Section IV(b)]. IV. Decomposition of Multiport Capacitors (a) Linear multiport capacitors The decomposition of a linear multiport capacitor is completely analogous to the decomposition of a MP R, because the constitutive matrix C of a MP C is also real, FIG. 10. Decomposition 264 of a modulated multiport resistor. Journal of the Franklin Institute Pergamon Press Lfd Decomposition of Multiport Elements symmetric [due to Maxwell reciprocity or energy conservation definite [due to stability conditions (9)]. This means that c= v-AI/= (6)] and positive WTW or rather C-’ with D a diagonal = T*D-‘T matrix, and e,=/C-‘dg=/C-‘f,dt. Analogous to a linearized MP R a linearized MP C may have operating points where the constitutive matrix C is not positive-definite. Such cases also have to be studied individually and are excluded from this treatment. Figure 11(a) shows the congruence canonical decomposition of a linear n-port capacitor similar to the canonical form of a MP R [Fig. 7(c)]. We write the equations : e, = TTe, fi = Tf, e, = D-‘fi dt. s Hence e, = TT and because C is a constant D-‘Tf, dt matrix if the MP C is linear, e, = T’D-‘Tf, s dt = C-‘f, (2) T is constant, so dt s which confirms the vahdity of this cananical decanposiiion. The immediate canonical forms of 2- and 3-port Cs are presented in Fig. 1 l(b), analogous to Fig. 8(d). According to the thermodynamic bond graph (TBG) concept (6, lo), recently renamed (11) as the generalized bond graph (GBG) concept in order to prevent Vol. 318, No. 4, pp. 253-273, Prmted in Great Bntam October 1984 265 Peter C. Breedveld FIG. 11.(a) Congruence canonical decomposition of a linear multiport capacitor. Immediate canonical forms of linear 2- and 3-port capacitors. (‘4 confusion caused by a common misinterpretation of the adjective “thermodynamic”, only capacitors have to be studied. However, in case one wishes to use inertances, the decomposition of a linear MP I is, of course, completely “dualogous” to the decomposition of a MP C, i.e. the roles of effort and flow in the equations have to be interchanged. (b) Nonlinear multiport capacitors Equation (2) already indicates that the decomposition is less straightforward in the nonlinear case, i.e. in case T is dependent on the state variables ql. Time-variant “capacitors” are not considered because they would violate energy conservation which does not correspond with the fundamental concept of a capacitor. It is assumed that the constitutive equation of a nonlinear MP C can be written in the derivative form : C$e,=f,. 266 Journal of the Franklin I~st~ture Pergamon Press Ltd. Decomposition of Multiport Elements Now suppose that (D is a diagonal C = (Tp’(q,)DTmT(q,)) matrix) e, = TT(ql)eZ fi = T(q,)f, then C&(TTe,) = T-‘fi or TC$(T’e,) = fi. This can be written d (TCTT)dte, d = Dze, = f; (3) and {TC$(T’))e, = {D(T-‘);ldt(T’))% =f; with f* = fi+fl;. (5) Equation (3) characterizes a capacitor array with D as constitutive matrix and (4) a state modulated, nonlinear n-port element, which shows some similarity to the so-called “gyristor” which is always modulated (MGR). This element has been introduced by Allen (12) as an auxiliary multiport, occurring in transformations of storage elements over state-modulated transformers and extended to nonsymmetric cases in (13, 6). Such a transformation may be considered to be a composition, i.e. the MP TF is included in instead of extracted from the MP C. If the result of such a composition has to result in a nonlinear MP C only (the starting point of the decomposition), there has to be some gyristor-like MP which annihilates the gyristor after transformation. This inverse reasoning leads also to the conclusion that a decomposition of a nonlinear MP C generates a gyristor-like element. Equation (5) is the flow-balance of a O-junction array determined bye,. Hence Fig. 12 seems to be a possible decomposition of a nonlinear MP C. Suppose it is required that the energy stored in the MP C is equal to the energy stored in the unit capacitor array. Then the gyristor (MP MGR) is a non-energic, non-dissipative (“non-entropic”) MP and consequently is a MP MGY, characterized by an antisymmetric matrix (a symmetric part would produce or annihilate entropy) : T DT-~$(T’)= Vol. 318, No. 4, pp. 253-273, Printed m Great Britain October - 1984 > = -&T)T-ID 267 Peter C. Breedveld FIG. 12. Decomposition of a nonlinear multiport capacitor with the use of a modulated multiport gyristor. or T-‘D$(T-‘)T’+ $(T’)D=0 applying the chain rule ($(T’DF)P- $(PD) + ;(T-‘D)-T-‘$(D)=0 ,( ) which means that $(C)=T-‘$(D)T-? The condition that C is nonlinear : thus results in $DfO. This means that the capacitor array characterized by D has to be nonlinear, i.e. Dii = Dii(qi), in order to obtain a non-energic, non-entropic gyristor (gyrator) in the decomposition. The problem is that this result can not be obtained by mere matrix manipulations. Secondly the state-modulation of the MP MTF (which corresponds to a holonomic constraint) is internal and this shows that the decomposition is also not very informative, because every type of element can be constructed out of other elements with the use of internal modulation, as will be demonstrated in the next subsection. This, in fact, questions the terminology “element”. Taking also into consideration that the decomposition of a nonlinear MP C is complicated and that (causal) inversion leads to loops which can be unstable (sometimes the gain is positive, Fig. 13), this decomposition seems to bring neither computational nor conceptual advantages. It may be concluded that general rules for decomposition of nonlinear multiport storage elements cannot be found, and that the possibility of decomposition must be studied separately for every specific type of nonlinear MP C. 268 Journal of the Franklin Institute Pergamon Press Ltd. Decomposition FIG. 13. Loops in the decomposition (c) The ambiguity of a nonlinear causality. of decomposition multiport of Multiport capacitor Elements with derivative by internal modulation An elementary linear capacitor, for instance, can be replaced by a source internally modulated transformer [Fig. 14(a)] and an irreversible transducer replaced by an internally modulated transformer or gyrator, depending causality [Fig. 14(b)] (14). This shows that “elements” can be constructed out “elements” by internal modulation, such that the basic concepts would “basic” anymore. In order to exclude this ambiguity, decomposition by modulation cannot be allowed. This does not mean, however, that modulation should be rejected in all cases. V. Decomposition of Multiport and an can be on its of other not be internal internal Gyrators The multiport gyrator as defined in (6,4) (cf. Fig. 15) has a dualistic character : with respect to its constitutive relation it is the counterpart of the MP R [cf. Casimir’s extension of the Onsager reciprocity relations as discussed in (6)] and with respect to its “physical” characteristics (non-energic, non-entropic) it is the partial dual of the MP TF and may be considered to be a special form of the so-called generalized junction structure (GJS) (7, 15). Its decomposition can also be approached in two ways, both leading to a canonical form. (b) +RS+ = d s FIG. 14. (a) Construction of a l-port linear capacitor by means of internal modulation. Irreversible transducer as an internally modulated transformer or gyrator. Vol. 318, No. 4, pp. 253-273, Prmted in Great Britain October 1984 (b) 269 Peter C. Breedveld FIG. 15. Multiport gyrator. FIG. 16. Immediate decomposition modulated gyrator. The first approach to be discussed is similar to Each of the (42) (n - 1) independent elements of metric matrix is considered to be a constitutive connecting the ports corresponding to its indices. of a 3-port the decomposition of a MP TF. an n x m-dimensional, antisymparameter of a 2-port gyrator In terms of an algorithm : -the ith port of an n-port gyrator with constitutive matrix G is connected to all other ports j # i by a 2-port gyrator with transformation ratio G, -the “summation points” at the ports are l-junctions in case of effort-causality, i.e. : e = Gf and O-junctions in case of flow-causality, i.e. f=G-‘e. -if i <j the gyrator bond connected to the ith port has the same orientation as the bond representing the port, while for i > j the gyrator bond connected to the ith port is oppositely oriented to the bond representing the port. A decomposition according to this algorithm is minimal because the number of 2ports is equal to the number of independent matrix elements of G. Accordingly this decomposition is called the immediate canonicalform. No other conditions than the antisymmetry of G have been used, so the decomposition is also valid in case the gyrator is modulated (MP MGY). Applied to a (modulated) 3-port gyrator, the algorithm results in the well-known triangle structure (Fig. 16) which occurs, for instance, in case of 3-D rotating reference frames as the so-called Eulerian junction structure (EJS) (16, 17, 6). This decomposition shows that a general form of a 3-port gyrator always contains an elementary gyrator in an essential way, because it is not possible to eliminate all gyrators by partial dualization (6). Using the congruence properties of antisymmetric matrices (18), another decomposition rule can be found analogous to the decomposition of a MP R, which minimizes the number of gyrators t (congruence canonical form). Every real 270 Journalof the Franklin Institute Pergamon Press Ltd. Decomposition Iv I T I SGY ‘/o,& -’ Is matrix G is congruent B=diag{Si where :diag(. .,Si ,... Elements 1 ,G P.-z, FIG. 17. Congruence decomposition antisymmetric of Multiport of a multiport gyrator. matrix B : to a canonical ,..., Si ,..., S,,O ,..., O> Si = 1 0 1 [ -1 0’ Hence G = TTBT. It follows immediately that the rank of G, r = 2t, is even and that odd dimensional, antisymmetric matrices are always singular. Figure 17 shows the congruence decomposition, where the MP TF with constitutive matrix T can be decomposed according to Section II. The SGY is a 2-port unit gyrator, which has been called a “symplectic” gyrator because its constitutive matrix (SJ is called a symplectic matrix (19,15,6). Figure 18(a) shows the congruence canonical form of a 3-port GY, i.e. the SGY becomes a GY and only 2 TFs remain after junction structure manipulations. There seems to be no general rule to obtain the congruence canonical decomposition of an arbitrary n-port GY. (a) e, f, e, T/O +1 / ’ 9: ‘l\+ 9:GY =2 = gt,f, e % f, (b) t -@A = /cl 4 3 = gf + cgt,t, - gt, t*1 f* -gf3 -gt,f, + gf,f, 0 e ’ f, +, 2’ CGYg+;.GY GY-g+, e3 f, >I FIG. 18. (a) Congruence Vol. 318, No. 4, pp. 253-273, Prmted in Great Britain T i canonical October form of a 3-port gyrator. canonical form. (b) Corresponding immediate 1984 271 Peter C. Breedveld The congruence canonical form is useful to check if a multiport gyrator is “essential”, i.e. if its decomposition contains a 2-port gyrator in an essential way (15). However, if a MP GY is essential, the congruence decomposition is not suited for computational purposes, because it then contains a positive and a negative resistive term in the constitutive equation which cancel each other. Numerical inaccuracies may destroy this cancelling effect with the risk of unnecessary instabilities. For instance, iffy in Fig. 18(a) is very large with respect to tlf2, the term gtlfi will be lost in the equation for e3, but - tzfi in the equation for e, remains, and consequently a negative resistive term -t, t2gf2 in the equation for e2. Hence, a port which “sees itself causally” through a gyrator has to be avoided and replaced by the immediate canonical form [Fig. 18(b)]. VI. Conclusion The decomposition rules presented for (bond graph) multiport elements are partly well-known and partly new, but it was judged useful to present an extensive enumeration and formalization. Decomposition can be a very powerful tool in the analysis and design of engineering systems containing multiport elements, such as transducers, because it provides a deeper insight into the characteristic properties of a dynamic system in terms of accepted concepts. This is especially true if certain properties which appear as separate elements in the decomposition have to be optimized. Since it would appear to be possible to implement decomposition algorithms as formulated in this paper in simulation software, this could be an important step towards more efficient computer-aided modelling and design. References (1) J. W. Meerman, “THTSIM, software for the simulation of continuous dynamic systems on small and very small computer systems”, Int. J. Modelling Simulation, Vol. I, No. 1, pp. 52-56, 1981. (2) D. C. Karnopp and R. C. Rosenberg, “System Dynamics : A Unified Approach”, Wiley, New York, 1975. (3) V. D. Gebben, “Bond graph bibliography”, J. Franklin Inst., Vol. 308, pp. 361-369,1979; and A. M. Bos and P. C. Breedveld, “1983 update of the bond graph bibliography”, J. Franklin Inst., in press. (4) P. C. Breedveld, “Proposition for an unambiguous vector bond graph notation”, Trans. ASME, J. Dynamic Syst. Measurement Control, Vol. 104, No. 3, pp. 267-270, Sept. 1982. (5) L. S. Bonderson, “Vector bond graphs applied to one-dimensional distributed systems”, Trans. ASME, J. Dynamic Syst. Measurement Control, Vol. 97, No. 1, pp. 75-82,197s. (6) P. C. Breedveld, “Thermodynamic bond graphs and the problem of thermal inertance”, J. Franklin Inst., Vol. 314, No. 1, pp. 15-40, July 1982. (7) R. C. Rosenberg, “Essential gyrators and reciprocity in network structures”, J. Franklin Inst., Vol. 308, pp. 343-352, 1979. (8) L. Onsager, “Reciprocal relations in irreversible processes-I”, Phys. Rev., Vol. 37, pp. 405426, 1931; and “Reciprocal relations in irreversible processes-II”, Phys. Rev., Vol. 37, pp. 2265-2279, 1931. 272 Journal of the Franklin Institute Pergamon Press Ltd. Decomposition of Multiport Elements (9) H. B. Callen, “Thermodynamics”, Wiley, New York, 1960. (10) P. C. Breedveld. “The thermodynamic bond graph concept applied to a flapper-nozzle valve”, Proc. 10th IMACS World Congress System Simulation and Scientijic Computation, Vol. 3, pp. 395-397,8-13 Aug. 1982. (11) P. C. Breedveld, “A bond graph algorithm to determine the equilibrium state of a system”, J. Franklin Inst., Vol. 318, pp. 71-75, 1984. (12) R. R. Allen, “Multiport representation of inertia properties of kinematic mechanisms”, J. Franklin Inst., Vol. 308, No. 3, pp. 235-253, 1979. (13) P. C. Breedveld, “Comment on multiport representations of inertia properties of kinematic mechanisms”, J. Franklin Inst., Vol. 309, No. 6, pp. 491492, 1980. (14) H. M. Paynter, “Multiport dissipators as ideal work-into-heat converters”, in Proc. 10th Annual Pittsburgh Co& (Edited by W. G. Vogt and M. H. Mickle), pp. 1843-1845, Instrument Society of America, 25527 April 1979. (15) P. C. Breedveld, “Essential gyrators and equivalence rules for 3-port junction structures”, J. Franklin Inst., Vol. 318, pp. 77-89, 1984. (16) D. C. Karnopp, “The energetic structure of multibody dynamic systems”, in “System Structures in Engineering”, (Edited by 0. Bjorke and 0. I. Franksen), pp. 317-342, Tapir, 1978. (17) M. J. L. Tiernego, “Bond graph modelling and simulation techniques applied to a three axis driven pendulum”, Int. J. ModelZing Simulation, Vol. 1, No. 1, pp. 62-66, 1981. (18) F. Ayres, Jr., “Theory and Problems of Matrices”, McGraw-Hill, New York, 1974. (19) P. C. Breedveld, “Thermodynamic bond graphs: A new synthesis”, Int. J. ModelZing Simulation, Vol. 1, No. 1, pp. 57-61, 1981. Vol. 318, No. 4, pp. 253-273, Printed in Great Britain October 1984 213